Unsupervised Multi-Source Domain Adaptation for Regression

This algorithm, working in the difficult unsupervised DA setting, itera- tively builds a combination of weak DA learners able to minimize both the source classification error and

This method aims at finding the optimal transport plan and training the classifier in the same time, the authors suggests that this method allows to consider both the data

We proved a new bound on this latter that involves source margin violation error, a novel convex alignment term given by a task-dependent variant of the Wasserstein distance between

Matrix

Pascal Germain, Amaury Habrard, François Laviolette, Emilie Morvant.. An Improvement to the Domain Adaptation Bound in a

– Only near.: models trained using only the nearest domain (according to our distance measure) to the target one; – Only far.: models trained using only the farthest domain

The introduced ap- proach, Joint Class Proportion and Optimal Transport (JCPOT), performs multi-source adaptation and target shift correction simul- taneously by learning the

The general comparison for CaffeNet features, SURF and GoogleNet features is given in Table 2. As before, we observe an important difference between taking the first most similar